A Friendly Introduction to the World's Oldest Living Medical Vocabulary

If there's one medical code you've probably seen on a bill, an insurance form, or a doctor's note, it's an ICD code. Something like E11.9 or J45.909 — a letter followed by some digits, occasionally with a decimal point. Those codes are everywhere. They drive billing, public health surveillance, mortality statistics, clinical research, and increasingly, AI-powered risk adjustment.

But ICD is more than a single code set. It's a family of code sets — international, national, diagnostic, procedural — with a 130-year history and at least three different versions in active use in the U.S. today. It can be confusing, even for people who work with it every day.

This post is a friendly tour through the whole ICD family: where it came from, how it's structured, who maintains it, who uses it, how to get it, and how it connects to the rest of the healthcare data ecosystem. By the end, the alphabet soup should feel a lot less intimidating.

Let's get into it.

What is ICD, really?

ICD stands for the International Classification of Diseases. At its heart, it's a system for classifying every disease, injury, symptom, sign, abnormal finding, and external cause of morbidity or mortality into a structured taxonomy of codes.

The World Health Organization (WHO) maintains the international version. Many countries — including the United States — produce their own clinical modifications that extend the international standard for local use. The U.S. version, ICD-10-CM, has roughly 70,000 diagnosis codes; the WHO international version has around 14,000. The U.S. wanted more granularity, so it built more granularity.

ICD codes show up in three big places:

1. On claims — every diagnosis billed to an insurer is encoded in ICD-10-CM.

2. On death certificates — every cause of death is encoded in ICD (for U.S. mortality, currently ICD-10).

3. In public health and research — disease surveillance, outbreak tracking, registries, and epidemiological studies all use ICD.

That triple role — billing, mortality, public health — is what makes ICD uniquely consequential.

A brief history: from a 19th-century statistical project to a global standard

ICD is the oldest medical vocabulary still in active use, and its history is genuinely fascinating. Pull up a chair.

1893 — The Bertillon Classification. A French physician and statistician named Jacques Bertillon presented the International List of Causes of Death at the International Statistical Institute meeting in Chicago. The goal: compare mortality data across countries, which was impossible when every country used its own disease vocabulary. Bertillon's list was adopted by several countries within a few years.

1900 — First international revision. The French government convened the first International Conference on revising the Bertillon Classification. The convention established a practice that continues to this day: revising the classification roughly every decade.

1900–1948 — ICD-1 through ICD-5. Revisions in 1900, 1909, 1920, 1929, and 1938 — each refining and expanding the list of causes of death. Originally the classification was only for mortality.

1948 — WHO takes over with ICD-6. When the World Health Organization was established after World War II, it took over stewardship of the classification. ICD-6 was a watershed: for the first time, the classification covered morbidity (diseases people lived with) in addition to mortality. ICD was no longer just for death certificates.

1955, 1965, 1975 — ICD-7, ICD-8, ICD-9. The revisions continued on roughly a decade-long cycle. ICD-9 (1975) became a major workhorse, used for both morbidity and mortality across most of the world.

1979 — ICD-9-CM in the United States. The U.S. National Center for Health Statistics (NCHS) created ICD-9-CM — a Clinical Modification of WHO ICD-9 — adding the extra specificity U.S. clinicians and payers wanted. ICD-9-CM became the U.S. standard for both diagnosis coding (Volumes 1–2) and inpatient procedure coding (Volume 3). It lasted 36 years.

1990 / 1994 — WHO releases ICD-10. WHO endorsed ICD-10 in 1990 and published it in 1994. Most of the world adopted it within a few years for mortality coding.

2015 — The U.S. finally transitions to ICD-10. After multiple delays stretching back to the late 1990s, the U.S. moved from ICD-9-CM to ICD-10-CM (diagnoses) and ICD-10-PCS (inpatient procedures) on October 1, 2015. This was one of the largest IT and operational transitions in the history of American healthcare — billions of dollars of preparation, training, and software updates.

2019 — WHO releases ICD-11. The 72nd World Health Assembly adopted ICD-11 in May 2019, with a planned global effective date of January 1, 2022. ICD-11 is digital-first, supports FHIR APIs, includes a semantic network, and introduces post-coordination — closer to a modern ontology than its predecessors.

Today. Most of the world is somewhere between ICD-10 and ICD-11 for morbidity coding. The U.S., however, is firmly on ICD-10-CM/PCS for billing and is not expected to transition to ICD-11 for at least several more years. We'll come back to this.

The ICD family in the U.S.: three (or four) variants

Here's where many newcomers get confused. When Americans say "ICD," they could mean any of several different things. Let's untangle them.

ICD-10-CM (Clinical Modification) is the U.S. diagnosis code set. Maintained by the CDC's National Center for Health Statistics (NCHS) with coordination from CMS. Roughly 70,000 codes. Required for all HIPAA-covered transactions since October 1, 2015. This is what you see on professional claims, outpatient hospital claims, and as diagnosis codes on inpatient hospital claims.

ICD-10-PCS (Procedure Coding System) is the U.S. inpatient hospital procedure code set. Maintained by CMS (developed originally under contract with 3M). About 78,000 codes. Used only for inpatient procedures billed on the UB-04 / 837I institutional claim. It replaced ICD-9-CM Volume 3 in 2015.

ICD-10 (WHO) is the international diagnosis classification. The U.S. uses it for mortality coding (death certificates) at NCHS — not for billing. About 14,000 codes.

ICD-11 (WHO) is the newest international version. WHO has released it, but the U.S. has not adopted it for billing or morbidity reporting. Mortality coding may transition first.

If you're working with U.S. claims data, you are almost certainly working with ICD-10-CM and possibly ICD-10-PCS. The rest of this post focuses on those, with detours into ICD-10 and ICD-11 where helpful.

How is ICD structured?

ICD's structure varies meaningfully across versions. Let's walk through them.

ICD-10-CM (diagnosis)

ICD-10-CM is a monohierarchical, alphanumeric classification. Each code lives in exactly one place in the hierarchy. The structure:

• Chapters — 22 of them, organized by body system or etiology (e.g., Chapter I: Certain Infectious and Parasitic Diseases; Chapter IX: Diseases of the Circulatory System; Chapter XX: External Causes of Morbidity)

• Blocks — groups of related conditions within a chapter

• Categories — three-character codes that name a condition family (e.g., J45 Asthma)

• Subcategories and codes — three to seven characters that add specificity

Codes are 3 to 7 characters. The first character is a letter (with some chapters using specific ranges — A–B for infectious diseases, C–D for neoplasms, E for endocrine, F for mental health, and so on). The second character is numeric. Characters 3–7 are alphanumeric.

A few features make ICD-10-CM uniquely expressive:

• The 7th character extension captures episode of care (initial, subsequent, sequela), laterality (left, right, bilateral), or trimester (for obstetric codes). For example, S52.501A is a closed fracture of the right radius, initial encounter.

• Combination codes capture etiology and manifestation in one code. For example, E11.21 is "Type 2 diabetes mellitus with diabetic nephropathy" — a single code expressing both the underlying disease and the complication.

• The "placeholder X" keeps codes the right length when a position is unfilled. For example, T36.0X1A uses X as a placeholder so the meaningful 7th character (A) lands in the right spot.

ICD-10-CM also distinguishes between the Tabular List (the codes themselves, in hierarchical order) and the Alphabetic Index (look up by clinical term to find a code). Both are official.

ICD-10-PCS (inpatient procedures)

ICD-10-PCS is structured completely differently from ICD-10-CM, and this surprises many people new to it. PCS is a multi-axial, post-coordinated code system. Every code is exactly 7 characters, and each character position is a defined axis:

1. Section — Medical and Surgical, Obstetrics, Imaging, Mental Health, etc.

2. Body System — Central Nervous System, Heart and Great Vessels, etc.

3. Root Operation — what the surgeon actually did (Excision, Resection, Bypass, Repair, etc.)

4. Body Part — what tissue or organ was operated on

5. Approach — Open, Percutaneous, Percutaneous Endoscopic, etc.

6. Device — any device left in (synthetic graft, drainage device, no device)

7. Qualifier — any additional clarification

There are 31 root operations, each with a precise, distinct definition. The whole system is designed to be constructed rather than looked up. A coder reads the operative report and builds the 7-character code by walking through the PCS Tables, choosing the appropriate value at each axis.

This makes ICD-10-PCS extraordinarily granular — it can describe procedures that didn't exist when the system was designed, because the axes are general enough to combine in new ways.

ICD-11

ICD-11 introduces a structure closer to a modern ontology. It has two main components:

• The Foundation Component is a polyhierarchical semantic network where any entity can have multiple parents.

• Linearizations are tabular projections of the Foundation tailored for specific uses (mortality and morbidity statistics, primary care, etc.).

ICD-11 also supports post-coordination clusters — you can combine a "stem code" with one or more "extension codes" to express specificity that would have required a unique code in ICD-10. It has stable URIs for every entity and was designed for digital use from the start.

Where does ICD live, and how often is it updated?

This is one of those topics where the answer depends on which version of ICD you're asking about. Let's go through them.

ICD-10-CM updates

ICD-10-CM is updated annually, effective October 1 of each year (aligned with the federal fiscal year). New codes for FY 2026 took effect on October 1, 2025.

Since FY 2022, CMS has also recognized a secondary April 1 update window for urgent new codes — typically new-technology add-on payments and public-health emergency needs. For example, several new COVID-19 codes were released off-cycle.

Updates are decided by the ICD-10 Coordination and Maintenance Committee, a federal interdepartmental committee co-chaired by NCHS and CMS. The committee holds public spring and fall meetings, considers code-change requests, and publishes proposals for comment.

The authoritative source for ICD-10-CM is published in two places:

• CDC NCHS at cdc.gov/nchs/icd/icd-10-cm/ — the editorial home, with guidelines, addenda, and errata

• CMS at cms.gov/medicare/coding-billing/icd-10-codes — the annual fiscal-year files

Both publish the same code set; CDC owns the editorial content and CMS owns the operational distribution. Files come as XML (the canonical machine-readable form), fixed-width text "order files," PDFs of the official guidelines, and Excel companion tables.

ICD-10-PCS updates

ICD-10-PCS is also updated annually on October 1, with the same April 1 secondary cycle since FY 2022. CMS owns the maintenance. New code requests go through MEARIS (Medicare Electronic Application Request Information System) at mearis.cms.gov.

Files include the tables (which is how PCS codes are constructed), the alphabetic index, definitions, an order file, and a reference manual. All free from CMS.

ICD-11 updates

WHO updates ICD-11 annually, distributed through the WHO ICD platform at icd.who.int. Browser, coding tool, and API are all free with a developer account.

What about ICD-11 in the U.S.?

Here's the question everyone in healthcare IT eventually asks: when will the U.S. switch to ICD-11?

The honest answer is: not soon. The National Committee on Vital and Health Statistics (NCVHS) created an ICD-11 Workgroup in 2023 to study the question, with Phase I evaluation running 2023–2025. The most credible projections:

• Mortality coding by NCHS: probably 2025–2027

• Morbidity / billing: not before 2027–2029, and many experts publicly expect a 10–15 year horizon

Adoption is hard because only about 23.5% of ICD-10 codes have direct ICD-11 equivalents, and ICD-11's stem-extension cluster model is a significant adjudication change for payers. If you see trade-press articles claiming "ICD-11 in 2025," treat them as speculative.

For practical product planning, assume ICD-10-CM/PCS will be the U.S. standard for at least the next 5–7 years, with optionality to add ICD-11 for international and research customers.

Who uses ICD, and why?

The short list: pretty much everyone in U.S. healthcare.

Every HIPAA-covered entity is required to use ICD-10-CM on claims. That includes physicians, hospitals, payers (Medicare, Medicaid, commercial), clearinghouses, and HIEs. There is no opting out.

Hospitals use both ICD-10-CM (diagnoses) and ICD-10-PCS (inpatient procedures). The combination drives MS-DRG (Medicare Severity Diagnosis-Related Groups) inpatient payment under the Inpatient Prospective Payment System.

CMS uses ICD-10-CM for risk adjustment — the CMS-HCC model, currently in version 28 (fully operative for Payment Year 2026), maps ICD-10-CM diagnoses to Hierarchical Condition Categories that drive capitated payments to Medicare Advantage plans, ACOs, and ACA marketplace insurers. Risk adjustment is a multi-hundred-billion-dollar mechanism, and ICD-10-CM is its substrate.

Public health agencies — the CDC, state health departments — use ICD-10-CM for disease surveillance, outbreak tracking, and reportable conditions. They use WHO ICD-10 (and eventually ICD-11) for mortality.

Researchers use ICD codes extensively in claims databases, EHR-derived research datasets, and registries. Studies of disease prevalence, comorbidity, healthcare utilization, and outcomes almost all start with ICD codes.

Quality measurement programs (HEDIS, MIPS, CMS quality measures) define their populations using ICD-10-CM diagnoses combined with CPT procedures.

Payers use ICD-10-CM for medical necessity determinations, prior authorization, claims edits, and population health analytics.

Cancer registries use ICD-O-3 (a cancer-specific WHO classification), which sits alongside ICD-10-CM Chapter II for neoplasm reporting.

International health organizations use WHO ICD-10/ICD-11 for global health statistics, which is how WHO can publish things like global mortality rates from heart disease or tuberculosis.

How do you actually get ICD?

The licensing news here is mostly good — far simpler than CPT.

ICD-10-CM and ICD-10-PCS: free

Both ICD-10-CM and ICD-10-PCS are completely free in the U.S. Download from CMS or CDC NCHS. No license, no fees, no annual reports. You can redistribute, embed in software, build commercial products on top — go for it. This is one of the genuinely well-supported open code sets in U.S. healthcare.

The CDC ICD-10-CM landing page (cdc.gov/nchs/icd/icd-10-cm/) has the guidelines, addenda, errata, and full file packages. The CMS ICD-10 page has the same code files plus an interactive code lookup tool (icd10cmtool.cdc.gov).

WHO ICD-10 and ICD-11: free

WHO publishes ICD-10 and ICD-11 freely at icd.who.int. ICD-11 has a particularly nice browser, a coding tool, and a free API (OAuth2-based, free with developer registration). You can use them in academic and clinical applications without paying.

Third-party and open-source tooling

Because ICD is freely available, the third-party ecosystem is healthier than for CPT:

• icd10data.com — popular free web reference with a friendly interface and clinical context

• AAPC and AHIMA — coder credentialing organizations with reference materials and training

• TruCode, 3M 360 Encompass, Optum EncoderPro, Find-A-Code — commercial encoder products that combine ICD with CPT and other code sets

• Open-source libraries — icd (R package), icdpicr (R), python-icd10, icd10-cm (Node), and many others provide programmatic access

• VSAC value sets — the NLM's Value Set Authority Center publishes thousands of value sets built on ICD-10-CM (and other vocabularies) for quality measurement

• CMS Code Lookup tools — interactive web tools for ICD-10-CM and ICD-10-PCS

For ICD-11 specifically, WHO offers an excellent free API and openly publishes the coding tool and browser. Several FHIR terminology servers (HAPI FHIR, Snowstorm, Ontoserver) support ICD-10-CM as a CodeSystem.

How does ICD play with other vocabularies?

ICD doesn't live alone. It interlocks with most of the other major U.S. healthcare vocabularies.

ICD-10-CM ↔ SNOMED CT

The most important clinical cross-walk. The NLM publishes the SNOMED CT to ICD-10-CM Map as part of the U.S. Edition of SNOMED CT. It's rule-based and supports an algorithm called I-MAGIC that uses context (age, sex, comorbidities) to pick the most appropriate ICD-10-CM code for a given clinical SNOMED CT concept.

This map is what makes it possible for an EHR to capture a clinically rich SNOMED CT concept on the problem list and translate it to an ICD-10-CM code for billing. It is not strictly 1:1 — many SNOMED CT concepts map to multiple ICD-10-CM codes depending on context.

ICD-9-CM ↔ ICD-10-CM/PCS: the GEMs

When the U.S. transitioned from ICD-9 to ICD-10 in 2015, CMS published General Equivalence Mappings (GEMs) — bidirectional cross-references between ICD-9-CM and ICD-10-CM/PCS. The GEMs are frozen now (they haven't been maintained since the transition wrapped up), but they're still essential for historical claims data and longitudinal research that crosses the 2015 boundary.

A key caveat: GEMs are not 1:1. Many ICD-9 codes map to multiple ICD-10 codes and vice versa. They were designed as a translation aid, not a precise crosswalk. Use them with awareness of their limitations.

ICD-10-CM → MS-DRG

Inpatient hospital claims combine ICD-10-CM diagnoses and ICD-10-PCS procedures, which together feed into the MS-DRG Grouper to assign each discharge to one of about 772 MS-DRGs (as of FY 2026). The MS-DRG determines the hospital's payment under the Inpatient Prospective Payment System. The relationship is published in the MS-DRG Definitions Manual by CMS.

ICD-10-CM → CMS-HCC

ICD-10-CM diagnoses are mapped to Hierarchical Condition Categories (HCCs) for Medicare Advantage risk adjustment. The current model is CMS-HCC V28 (fully operative for Payment Year 2026), which uses about 7,770 valid ICD-10-CM codes mapped into 115 HCCs. The mapping is published annually by CMS as part of the risk-adjustment model software.

This is one of the highest-stakes uses of ICD-10-CM in healthcare. RADV audits, AI-powered coding tools, and the entire Medicare Advantage payment ecosystem ride on this mapping.

ICD-10-CM → DSM-5

The American Psychiatric Association's DSM-5-TR uses ICD-10-CM codes (F-chapter, mental and behavioral disorders) as its billing substrate. When you see DSM-5 diagnostic codes, those are ICD-10-CM codes. DSM provides the clinical criteria; ICD-10-CM provides the billable code.

ICD-10-CM ↔ ICD-O-3

For cancer, the International Classification of Diseases for Oncology (ICD-O-3) is a WHO classification specifically for cancer registries. It pairs topography (anatomic site, aligned with ICD-10) with morphology (cell type and behavior). ICD-10-CM Chapter II (Neoplasms) and ICD-O-3 are designed to work together, with cross-references between them.

ICD-10-CM ↔ MedDRA

Pharmaceutical companies maintain proprietary cross-walks between MedDRA (the regulatory adverse-event terminology) and ICD-10-CM for hybrid claims-and-safety analyses. There's no official map, but in UMLS the two are linked through Concept Unique Identifiers.

ICD-11 ↔ SNOMED CT

WHO and SNOMED International are collaborating on a mapping between ICD-11 and SNOMED CT. It's still in progress, but the goal is to align the two most important international clinical vocabularies.

ICD inside UMLS

The NLM's Unified Medical Language System (UMLS) includes ICD-9-CM, ICD-10-CM, ICD-10-PCS, ICD-10 (WHO), ICD-O-3, and ICD-11 as source vocabularies. Each is linked to Concept Unique Identifiers that integrate them with roughly 190 other source vocabularies. For research-grade cross-vocabulary work, UMLS is the obvious starting point.

Wrapping up

ICD is the most consequential code system in healthcare you might never think about directly. Every claim, every death certificate, every public health statistic, every risk-adjustment calculation, every clinical research cohort — they all start with ICD codes.

If you're learning U.S. healthcare data, the most important things to internalize are:

1. ICD-10-CM is the U.S. diagnosis code set for billing — about 70,000 codes, maintained by CDC NCHS with CMS, updated October 1 each year. Free.

2. ICD-10-PCS is the U.S. inpatient procedure code set — about 78,000 codes, maintained by CMS, updated October 1. Free. Structured very differently from ICD-10-CM (constructed, not looked up).

3. WHO ICD-10 is used in the U.S. only for mortality coding at NCHS. Free.

4. ICD-11 is WHO's newest version. The U.S. has not adopted it for billing and probably won't for several more years.

5. Cross-walks to SNOMED CT, MS-DRG, CMS-HCC, DSM-5, ICD-O-3, and historical ICD-9-CM (via the frozen GEMs) make ICD the connective tissue of U.S. healthcare data.

ICD has survived 130 years of medical progress, two world wars, the rise of computing, and the slow shift from paper to digital because it does one thing extraordinarily well: it gives the world a shared vocabulary for talking about disease. That utility is why it's still going — and why understanding it is one of the highest-leverage things anyone working in healthcare data can do.

Welcome to the wonderful, complicated, surprisingly historical world of ICD.